Fake leads in your CRM: do they get sent to Facebook as conversions?

Half your leads are fake, and you're sending conversions back to Meta. Does the fake half go too?

It depends on two things: which door the lead came through, and what you send.

First, a distinction most people skip. "Fake lead" is actually four different animals. A bot that fills your form. A human who fills it with junk: a competitor, a student, someone testing. A duplicate of a real lead. And an internal test that somebody forgot to delete. They behave differently, and they get filtered by different things.

Now the two doors. The website door: a visitor lands, fills in the form, and the pixel or server reports a "lead". Whatever fills the form gets counted, bots included. The CRM door: a contact reaches a stage in the pipeline and a tool reports it. That door carries no bot check at all, so a fake contact that somebody moved to "booked" in a hurry is sent like a real one.

Here's a made-up month to show why this matters. 100 leads, of which 40 are fake. Send the "lead" event straight from the form and Meta records 100 conversions, 40 of them teaching it the wrong customer. Worse, it does what it's built to do: it goes looking for more traffic like those 40. A bot doesn't just cost you a click. It changes who the algorithm goes after next, and that effect compounds.

Now send booked instead. Most bots never book. Most junk humans never turn up. So the 40 mostly fall away without you doing anything clever. That's one of the quiet benefits of sending later stages instead of leads, and hardly anyone mentions it.

But it isn't a full defence, and I don't want to oversell it. Some junk does book. Some tests do get moved through stages. So the honest approach is layers. Filter bots at the website door, before the conversion is sent. Verify the lead at the form. Qualify honestly in the CRM before any stage that gets sent. Each layer catches something different.

A ten-minute test to see where you stand. Look at your last 50 leads and mark each one: answered the phone, didn't answer, obviously fake. If fewer than half answered, your ads are very likely learning from junk. If your fake rate is higher on one campaign than another, that's a targeting or placement problem, and it's worth knowing before you optimise anything.

Here's a cheap way to tell the four animals apart in your own leads. Bots often submit within seconds of the page loading, with pattern-like names and emails, and never answer. Junk humans give plausible details, but the phone is disconnected or the person says they never asked for anything. Duplicates show the same email or phone twice, often minutes apart. Tests carry names like "test" or your own team's emails.

Sort your last 50 leads into those four buckets, and you'll usually find one bucket is most of your problem. That tells you which layer to fix first. Bots point to the website door. Junk humans point to targeting and form friction. Duplicates point to your CRM rules. Tests point to a filter you can add in five minutes.

For reference, DataCops has an option to hold the ad conversion until the CRM confirms the lead, so a fake one never trains the campaign. CRM events carry no bot check of their own.